Detracking Autoencoding Conditional Generative Adversarial Network: Improved Generative Adversarial Network Method...

Area of Science:

• Computer Science
• Machine Learning
• Data Science

Background:

• Missing values in datasets are a common challenge due to data collection limitations and network issues.
• Current generative adversarial imputation methods struggle with applicability, latent categorical information, and balancing local/global data aspects.

Purpose of the Study:

• To propose a novel generative adversarial model, DTAE-CGAN, to address limitations in existing data imputation techniques.
• To enhance the model's ability to learn inter-sample correlations and utilize all available data information in incomplete datasets.

Main Methods:

• Developed DTAE-CGAN, a generative adversarial model incorporating detracking autoencoding and conditional labels.
• Trained and evaluated the model on six real-world datasets of varying sizes.

Main Results:

• DTAE-CGAN demonstrated superior imputation accuracy compared to four classic imputation baselines across all tested datasets.
• The model effectively learns inter-sample correlations and leverages complete data information.

Conclusions:

• DTAE-CGAN offers a significant improvement over existing methods for handling missing data.
• The proposed model provides a robust and accurate solution for data imputation in various applications.